Stop teaching people to prompt. Teach them to build.
Prompting is a tactic. Building an agent that runs part of your own job is the skill that sticks.
Shubham Chandra, Head of Curriculum8 min read
Key takeaways
- Prompting is to AI what typing is to writing. It's the input, not the skill.
- Non-technical professionals can build a working agent in one live session, from a starter written for their role, without writing code.
- Two moments convert skeptics: the time savings shock (2 hours to 10 minutes) and the ownership moment ("I built this").
- Tools handed down by IT get tried once and dropped within weeks. Tools people build for their own work stay open on Monday morning, because they were built for Monday morning's task.
The prompting trap
- a week per person on steps an agent could run, as a working assumption
- 2.5 hrs
- a week per person on steps an agent could run, as a working assumption
- Our estimate; put in your own number
- per person per year at that rate
- $9,750
- per person per year at that rate
- 2.5 hrs × $75/hr fully loaded × 52 weeks
Everyone is teaching prompting. Workshops, webinar series and certification programs have made it the default answer to "how do we upskill on AI."
Prompting is to AI what typing is to writing. No company would send a team to a typing workshop and call it a communications program, yet most prompt courses deliver exactly that: a faster way to talk to a tool, with no attention to whether the tool is pointed at the right problem.
Take a conservative assumption: one person spends 2.5 hours a week on steps an agent could run. At $75 an hour, fully loaded, that is $9,750 a year per person, and it scales with the team. Prompt workshops aren't capturing it because they aren't changing what people do with their working hours.
What building looks like for non-technical people
None of these require code. Each is an agent built from a starter written for that role and pointed at one recurring task.
Tell someone "you're going to learn to build AI tools" and they picture software engineering. Watching the instructor build one on a volunteer's real work, then building their own the same day without writing code, dissolves that. The thing they built on day one becomes the thing they show their manager.
- HR coordinator: Builds an agent that drafts job descriptions in the company's voice and tone
- Marketing lead: Builds one that turns campaign data into exec summaries for the Monday standup
- Ops manager: Builds one that generates weekly status reports from project notes, replacing a 90-minute Friday task
- Sales team: Connects their CRM to an AI workflow that drafts follow-up emails from call notes
- Finance team: Automates monthly variance commentary that used to mean pulling data from three different systems
Your team can already prompt. Can they build something that saves them two hours every week?
The two moments that convert skeptics
The first is the time savings shock. A task that took two hours now takes ten minutes, and the Friday report is done before the coffee gets cold. It shows on their faces.
The second is ownership. Watching a demo proves the instructor can do it. Building one proves they can, and "I built this" is the sentence that turns a skeptic.
These two moments explain a pattern. AI tools that IT rolls out with a company-wide email get tried once and left alone. Agents that people build for their own tasks get opened again, because the builder already knows what the next run is for.
Ownership is the difference, and a platform license has never produced it.
Why this changes the adoption math
People use the tools they built, and abandon the ones somebody else handed them, usually within a few weeks.
Open online courses finish about 3% of the people who start, according to an MIT and Harvard study of edX learners, while HBS Online reports an 85% completion rate for its cohort-based courses. Even so, completion only proves attendance. The result that matters is someone doing a piece of their job differently on Monday.
The programs that change behavior are the ones where people leave with something they built and will open again on Tuesday morning.
| Measure | Tools imposed from above | Tools built from within |
|---|---|---|
| Usage after launch | Tried once, then dropped | Opened again the next week |
| Adoption timeline | Excitement for 2-3 weeks, then abandonment | Usage grows as people refine their tools |
| Maintenance | IT owns it; users submit tickets | Builders iterate on their own |
| Knowledge retention | Forgot the training by Friday | Built something they use every day |